Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
The unrestricted Lemma 4 on published pp. 7–8 is false as stated. Its conclusion is valid in the following sufficient range. Let be an integer, , be real, and . For arbitrary integer there are and integers such that
No coprimality assumption on the individual steps is required. The application in theorem_2 establishes before using this replacement when . Its one-dimensional case uses a unit step directly. The unrestricted printed statement receives no full-proof credit.
Source: published PDF, Lemma 4, pp. 7–8. The restriction and ceiling argument below are compilation-supplied repairs, not an author-issued erratum.
Bears on. Problem 297.
Counterexample to the printed range
Take , , and . The first bound in (1) would be . But the only allowed multiplier is , and its first residue is odd, so every integer representative has absolute value at least 1. All printed unit assumptions hold. The box count in the printed pigeonhole argument cannot discard integer rounding in this range.
Proof
Put , and . For ,
At there is equality; for the ceiling is 1; for , . Since , at least one factor has strict inequality. Consequently .
Partition in coordinate into half-open intervals of equal length. Their length is at most . For , form the vector of least nonnegative residues of . Two of these vectors lie in the same product box. If their indices are , let be the nonzero residue of in , and let be the difference of the two representatives in coordinate . These differences have the required congruences and absolute values strictly less than the indicated interval lengths. This proves (1).